The Agent Skills Directory

[DATA_EXFILTRATION]: The skill reads local shell profile files (e.g., ~/.zshrc, ~/.bashrc) and .env files to discover existing environment variables used for API authentication.
Evidence: Grep operations in SKILL.md targeting various shell initialization and environment files.
Context: This behavior is used to detect pre-configured credentials; the skill explicitly warns against accepting pasted secrets to prevent them from entering the conversation history.
[INDIRECT_PROMPT_INJECTION]: The skill processes USD property metadata from an external API to generate a catalog used as a grounding source for an LLM query parser.
Ingestion points: Fetches property statistics from USD_SEARCH_API_URL/search/stats/usd_properties and search results from /search_hybrid.
Boundary markers: No explicit instruction-isolation markers are implemented in the generated YAML/Markdown artifacts; safety relies on the platform's query parser constraints.
Capability inventory: The skill utilizes shell execution (Bash), file system write access, and Python script execution.
Sanitization: references/build_catalog.py implements regex-based slugging for field names and performs heuristic type inference to structure the ingested data.
[EXTERNAL_DOWNLOADS]: The skill dynamically installs the pyyaml library from the official Python package registry at runtime.
Evidence: Use of uv run --with pyyaml in SKILL.md.
Context: pyyaml is a well-known, industry-standard library for YAML processing.

usd-property-catalog